* Joel Stanley <j...@jms.id.au>:
> On Thu, 3 Aug 2023 at 13:55, Helge Deller <del...@gmx.de> wrote:
> > > 143551 brk(NULL) = 0x0009b000
> > > 143551 brk(0x0009b8fc) = 0x0009b000
> >
> > I think the problem is the brk with 9b000 here.
> > It's not 64k aligned (=pages size of your ppc64le).
> >
> > Please try with this patch on top of Richard's series:
> >
> > > @@ -3229,7 +3208,8 @@ static void load_elf_image(const char *image_name,
> > > int image_fd,
> > > info->end_code = 0;
> > > info->start_data = -1;
> > > info->end_data = 0;
> > > - info->brk = .....
> > change that to become:
> > info->brk = HOST_PAGE_ALIGN(hiaddr);
>
> That stopped the crashing, and the binaries seem to run fine. I tested
> on two hosts: ppc64le (64K) and arm64 (16K).
Great!
That made re-read Akihiko's patch:
----
Author: Akihiko Odaki <akihiko.od...@daynix.com>
linux-user: Do not align brk with host page size
do_brk() minimizes calls into target_mmap() by aligning the address
with host page size, which is potentially larger than the target page
size. However, the current implementation of this optimization has two
bugs:
- The start of brk is rounded up with the host page size while brk
advertises an address aligned with the target page size as the
beginning of brk. This makes the beginning of brk unmapped.
----
this patch has wrong assumptions.
The start of brk always needs to be host page aligned.
It's not an optimization, but a requirement, since brk needs to be
located on a host-aligned page which may get different permissions
than the page before it (where code from the binary may be located).
I wonder if we need that patch at all.
Joel, could you give the patch below on top of git head (no other
patches applied) a spin?
(I just tested it here locally on a full range of linux-user chroots)
I think this is ALL what's needed for git head to fix the static binary
issues, has a nice preparation for Richard's ELF_ET_DYN_BASE patches.
If it does, it replaces patches 1,2 & 4-6 from Richard's v7 patch
series.
Helge
diff --git a/linux-user/elfload.c b/linux-user/elfload.c
index 861ec07abc..88d9e4056e 100644
--- a/linux-user/elfload.c
+++ b/linux-user/elfload.c
@@ -3021,8 +3021,10 @@ static void load_elf_image(const char *image_name, int
image_fd,
struct elfhdr *ehdr = (struct elfhdr *)bprm_buf;
struct elf_phdr *phdr;
abi_ulong load_addr, load_bias, loaddr, hiaddr, error;
+ unsigned long load_offset = 0;
int i, retval, prot_exec;
Error *err = NULL;
+ bool is_main_executable;
/* First of all, some simple consistency checks */
if (!elf_check_ident(ehdr)) {
@@ -3106,28 +3108,8 @@ static void load_elf_image(const char *image_name, int
image_fd,
}
}
- if (pinterp_name != NULL) {
- /*
- * This is the main executable.
- *
- * Reserve extra space for brk.
- * We hold on to this space while placing the interpreter
- * and the stack, lest they be placed immediately after
- * the data segment and block allocation from the brk.
- *
- * 16MB is chosen as "large enough" without being so large as
- * to allow the result to not fit with a 32-bit guest on a
- * 32-bit host. However some 64 bit guests (e.g. s390x)
- * attempt to place their heap further ahead and currently
- * nothing stops them smashing into QEMUs address space.
- */
-#if TARGET_LONG_BITS == 64
- info->reserve_brk = 32 * MiB;
-#else
- info->reserve_brk = 16 * MiB;
-#endif
- hiaddr += info->reserve_brk;
-
+ is_main_executable = (pinterp_name != NULL);
+ if (is_main_executable) {
if (ehdr->e_type == ET_EXEC) {
/*
* Make sure that the low address does not conflict with
@@ -3136,10 +3118,11 @@ static void load_elf_image(const char *image_name, int
image_fd,
probe_guest_base(image_name, loaddr, hiaddr);
} else {
/*
- * The binary is dynamic, but we still need to
+ * The binary is dynamic (pie-executabe), but we still need to
* select guest_base. In this case we pass a size.
*/
probe_guest_base(image_name, 0, hiaddr - loaddr);
+ load_offset = 0 /* TODO: should be ELF_ET_DYN_BASE */;
}
}
@@ -3157,9 +3140,9 @@ static void load_elf_image(const char *image_name, int
image_fd,
* In both cases, we will overwrite pages in this range with mappings
* from the executable.
*/
- load_addr = target_mmap(loaddr, (size_t)hiaddr - loaddr + 1, PROT_NONE,
+ load_addr = target_mmap(loaddr + load_offset, (size_t)hiaddr - loaddr + 1,
PROT_NONE,
MAP_PRIVATE | MAP_ANON | MAP_NORESERVE |
- (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0),
+ (is_main_executable ? MAP_FIXED : 0),
-1, 0);
if (load_addr == -1) {
goto exit_mmap;
@@ -3194,7 +3177,8 @@ static void load_elf_image(const char *image_name, int
image_fd,
info->end_code = 0;
info->start_data = -1;
info->end_data = 0;
- info->brk = 0;
+ /* possible start for brk is behind all sections of this ELF file. */
+ info->brk = HOST_PAGE_ALIGN(load_offset + hiaddr);
info->elf_flags = ehdr->e_flags;
prot_exec = PROT_EXEC;
@@ -3288,9 +3272,6 @@ static void load_elf_image(const char *image_name, int
image_fd,
info->end_data = vaddr_ef;
}
}
- if (vaddr_em > info->brk) {
- info->brk = vaddr_em;
- }
#ifdef TARGET_MIPS
} else if (eppnt->p_type == PT_MIPS_ABIFLAGS) {
Mips_elf_abiflags_v0 abiflags;
@@ -3618,6 +3599,15 @@ int load_elf_binary(struct linux_binprm *bprm, struct
image_info *info)
if (elf_interpreter) {
load_elf_interp(elf_interpreter, &interp_info, bprm->buf);
+ /*
+ * Use brk address of interpreter if it was loaded above the
+ * executable and leaves less than 16 MB for heap.
+ * This happens e.g. with static binaries on armhf.
+ */
+ if (interp_info.brk > info->brk &&
+ interp_info.load_bias - info->brk < 16 * MiB) {
+ info->brk = interp_info.brk;
+ }
/* If the program interpreter is one of these two, then assume
an iBCS2 image. Otherwise assume a native linux image. */
@@ -3672,17 +3662,6 @@ int load_elf_binary(struct linux_binprm *bprm, struct
image_info *info)
bprm->core_dump = &elf_core_dump;
#endif
- /*
- * If we reserved extra space for brk, release it now.
- * The implementation of do_brk in syscalls.c expects to be able
- * to mmap pages in this space.
- */
- if (info->reserve_brk) {
- abi_ulong start_brk = HOST_PAGE_ALIGN(info->brk);
- abi_ulong end_brk = HOST_PAGE_ALIGN(info->brk + info->reserve_brk);
- target_munmap(start_brk, end_brk - start_brk);
- }
-
return 0;
}
